Abstract

Introduction

Patient survival for most cancers has increased dramatically with earlier diagnosis, improved treatments, and faster and more effective management of side effects [1]. Novel cancer therapies include medications that augment elements of the immune response to directly target and kill cancer cells or to replace and boost immune system functions; these collectively are referred to as cancer immunotherapy [2]. Commensurate with increased survival from new standards of care are increased efforts to understand and predict each individual’s response to and toxicity from these therapies [2]. For reasons not fully understood, variability exists in both efficacy and toxicity-related complications among patients receiving immune therapies [3]. Recently, gut microbiome and immune profiles have been identified as contributing factors to this inter-individual variability (reviewed in [4, 5, 6, 7]).

Microorganisms inhabit our skin and respiratory and gastrointestinal tracts and comprise a large portion of our overall cellular, metabolic, and genetic mass [8]. The organisms that comprise this community are collectively referred to as the microbiota while their collective genetic material is termed the microbiome [9]. The intimate relationship among microbiota composition, metabolic function, and the development and regulation of the immune system has been correlated with entities as disparate as inflammatory bowel disease, autism, asthma, and obesity [10, 11, 12, 13, 14]. Several microbes directly inflame tissues, and certain viruses and bacteria have been well described as direct causes of cancer, for example, Epstein-Barr virus (EBV) with Burkitt’s and Hodgkin’s lymphoma, human papilloma viruses (HPV) with cervical, head, neck, and anal cancers, and Helicobacter pylori with gastric cancers, among others [15]. Given the connection between gut microbiota and inflammatory diseases, researchers are exploring whether differences in the human microbiome have the potential to impact individual responses to cancer immune therapies (additionally reviewed in [4, 5, 6, 7]). Further, recent advances in sequencing technology have provided fine granular detail of a potential tumor microbiota in certain cancers [16, 17, 18]. With decreasing metagenomic sequencing costs, large scale observational studies and placebo-controlled trials can now be analyzed for longitudinal changes in microbial structure and gene content [19]. These techniques have revealed correlations between the microbiome and cancer immunotherapy outcomes in pre-clinical animal models and human clinical trials [20].

In this review, we first discuss methodologies for characterizing the microbiome, including technical limitations and challenges. Next, we describe ways in which the commensal gastrointestinal microbiota has been shown to interact directly with the immune system and how these interactions influence oncogenesis. We then focus on interactions among cancer immunotherapies and the microbiome and ways in which the microbiome can be modulated using prebiotics, probiotics, postbiotics, antibiotics and fecal microbiota transplantation (FMT) to improve survival and limit toxicity.

Measuring the microbiome

Traditionally, clinical assessment of bacterial populations is achieved by sampling the site of interest and culturing organisms in broth and agar plates, followed by phenotypic and molecular identification [21]. Although we have made significant advances in our ability to culture various organisms, a large proportion of human-associated microbes are time- and labor-intensive to grow [22]. Additionally, bacterial enrichment through culture can obscure abundance differences in the original sample because of differential and exclusionary growth [23]. Cell-free methods to measure bacterial presence and activity have become more popular and have allowed measurement of microbial populations at previously impossible scales, including in diagnostic microbiology [24]. Nucleic-acid based methods for detecting viruses and bacteria initially focused exclusively on their presence or absence [25], but recent advances have incorporated relative abundance measures to determine bacterial presence in clinical samples quantitatively to better understand the burden of disease [26].

The first step in determining microbial colonization of a body site by sequencing is proper sampling with sterile technique and preservation from nucleases [27]. Once a sample is obtained, genomic DNA is extracted using phenol-chloroform or commercially-available kits with yield and purity dependent on sample type [28, 29] (Figure 1A). The 16S ribosomal RNA (rRNA) gene is highly conserved and is nearly universally present in bacteria, and its sequence can be used to distinguish bacterial genera or species [30]. 16S rRNA gene amplicon sequencing is a cheap, quick, and reliable method for determining bacterial taxonomy after computational processing with publicly-available tools such as QIIME 2 or DADA2 [31, 32] (Figure 1B). Although 16S rRNA gene sequencing is a useful tool, it has several limitations. Precision is often limited to the genus level, especially when assigning operational taxonomic units (OTUs) in lieu of amplicon sequence variants (ASVs), which are preferred due to enhanced specificity [33, 34]. OTU-picking methods cluster reads that differ by less than 3% from a standard, but often group several species into a common genus. For example, OTU methods would assign the genus Neisseria to both pathogenic (e.g. Neisseria gonorrhea and N. meningitidis) and commensal Neisseria spp., which is clearly an important clinical distinction depending on the sample type [34]. Therefore, methods that assign ASVs to each sequence are preferred, thereby allowing resolution of single nucleotide polymorphisms (SNPs) and strain-level differences [34]. Further, 16S rRNA gene sequencing is limited to genomic DNA, does not account for variable copy number within bacterial genera, which could artificially increase relative abundance[35], and ignores plasmids and genomic DNA that encode antibiotic resistance, toxin production, and other virulence factors of clinical importance.

Open in a separate window

Figure 1.

Schematic for metagenomic analysis.

A) The first step in microbiome analysis is selecting the sample, obtaining it with sterile technique, and preserving it with consideration of DNAses and RNAses. Samples can be obtained directly from tumors, skin, or fecal samples. DNA extraction via phenol chloroform or commercially available kits yields metagenomic DNA. B) The 16S rRNA gene can be amplified and sequenced. Alternatively, metagenomic DNA can be shotgun sequenced to yield DNA fragments of varying sizes. C) 16S rRNA reads can be computationally processed with publicly available pipelines yielding relative abundance of taxa per sample. Ordination methods such as PCA and PCoA are utilized to identify similarities and differences in diversity metrics (alpha, beta, gamma) between samples or between groups. Reads from metagenomic shotgun sequencing can be processed to yield the same information. In addition to taxonomy, shotgun reads can be processed and mapped to databases to determine functional pathways, assemble bacterial genomes, and understand the presence of viral or eukaryotic DNA depending on appropriate sequencing depth. Created with BioRender.

Alternatively, DNA from all organisms in the sample, collectively referred to as the metagenome, can be shotgun sequenced using commercial technology, and the taxonomy can be obtained using clade-specific markers with bioinformatic tools such as Kraken or MetaPhlAn2 [36, 37, 38] (Figure 1C). Additionally, shotgun metagenomic sequencing can be used to infer whole community microbial functional information and annotate genes that confer antibiotic resistance, produce toxins, or provide metabolic pathway information via comparison to known databases, assuming appropriate sequencing depth [27, 39, 40, 41] (Figure 1C). Metagenomic sequencing can also identify non-bacterial components of the microbiota, such as fungi, viruses, and bacteriophages, which are gaining attention for their roles in the microbiome and human health [42, 43]. Therefore, desired analyses necessarily determine whether 16S rRNA or metagenomic shotgun sequencing is appropriate to profile the microbiota or microbiome, respectively.

After the sequencing data is obtained, the first question asked is ‘who is there?’ Both 16S rRNA and shotgun metagenomic sequencing can answer this question (Figure 1C). Data from sequencing analysis can be visualized in terms of relative abundance of taxa within a sample or as a similarity or dissimilarity ordination plot based on measures of alpha and beta diversity (Figure 1C). Alpha diversity measurements describe intra-sample diversity: measurements include Shannon diversity and species richness. Beta diversity measures are pairwise comparisons of dissimilarity between all sample pairs and are typically viewed as an ordination plot with principal coordinate analysis (PCoA). Measures of beta diversity include Bray-Curtis, binary-Jaccard, and weighted UniFrac (Figure 1C) (reviewed extensively in [44]). In addition to overall assessments of similarity and dissimilarity between samples, these same measures can be used to examine differences in microbial gene content (e.g. functional pathways and resistance genes) from metagenomic shotgun sequencing (Figure 1C) (Reviewed in [27]). Although shotgun metagenomic sequencing is more expensive than 16S rRNA gene sequencing per sample, when performed at a depth of 2.5 – 5 million reads per sample, the validation of shallow shotgun sequencing with as few as 500,000 reads per sample increases the affordability of this method for large datasets [33]. Thus, in designing any study interrogating the microbiome, it is vital to determine the desired downstream analyses so that sampling schema and sequencing depth can be planned appropriately.

After sequencing and characterizing a microbial community, correlations can be discerned by connecting individual taxa, groups of taxa, or overall trends in community composition (e.g. diversity) to host metadata. However, this observation-driven strategy is frequently insufficient to draw concrete conclusions or to elucidate underlying mechanisms in the absence of manipulation or recapitulation in animal models. To achieve mechanistic understanding, microbiota-humanized mice are produced by gavaging germ-free or antibiotic-treated, conventional mice with stool samples from healthy humans, patient cohorts, or pre-defined consortia [45, 46]. Different microbes colonize mice with varying success, necessitating quality checks to ensure the input is comparable to the output community with at least 65% uptake at the phylum level, though the definition of “successful” microbiota-colonization depends on the richness and taxonomic abundance of the input community [45, 46, 47].

Designing in vivo microbiome-related experiments is accompanied by its own set of challenges and considerations. If humanized mice are used, will they be derived from germ-free mice or antibiotic-treated mice? Germ-free mice are free of all currently detectable microorganisms and allow complete recolonization, but they are expensive to maintain, have severely underdeveloped immune systems that are not always restorable, and must be individually derived for unique genotypes [48, 49]. If antibiotic-treated mice are used, which antibiotic(s) will be used to deplete the conventional mouse microbiota, and what confounding effects may they produce? Individual antibiotics and antibiotic cocktails affect different microbial populations and have varying efficacy and tolerability in mice [48]. Even seemingly arbitrary decisions require due diligence. Considering the appropriate vendor and housing conditions for isogenic conventional mice is vital because microbiota compositions differ among mice that are obtained from different suppliers and that are housed in different facilities [50]. Once mice are obtained and appropriately humanized, it is important to ensure they are consistently colonized and that their microbiota are stably established over time with minimal strain dropout [45, 46, 47]. For short experiments, this can be done by cohousing animals, wherein they will engage in coprophagy and exchange microbiota content [51]. For longer studies, littermate crossing can aid in microbiome stability between cages [51, 52]. All these factors must be taken into consideration not only when designing experiments, but also when interpreting results, reading the literature, and extrapolating conclusions about the role of the microbiota in human health from animal studies.

An additional concern, especially when considering environments previously presumed to be sterile or when tissue is isolated, is to include proper negative controls. Sequencing contamination can strongly influence results; therefore, researchers should include the extraction kit alone with no sample added or a skin swab control obtained at the same time and processed in the same way as the tumor sample. With proper controls, several bioinformatic tools exist to minimize the impact of contamination [53, 54]. Because of the advances in sequencing and computational techniques, it is vital to consider contamination when designing experiments to examine relationships between cancer and the local or gut microbiota. Indeed, Robinson et al. analyzed a large public repository, Cancer Genome Atlas, for microbial DNA and identified numerous skin contaminant taxa such as Staphylococcus epidermidis and Cutibacterium acnes. They also identified sequencing center-specific contaminants, reinforcing the importance of including appropriate negative controls [16]. To uncover associations between microbes and cancer, consideration of experimental design, including negative controls and sterile sample collection, is therefore vital.

Local tumor microbiota

Up to 15% of cancer malignancies worldwide can be attributed to infectious diseases [55]. Oncogenic viruses, such as EBV and HPV carry genes that directly interfere with cell cycle checkpoints and inhibit the activity of tumor suppressors, and thus can be considered initiators of cancer [56]. In contrast, bacteria tend to be considered cancer promoters; they contribute to oncogenesis by inducing inflammation and recruiting immune cells, thus increasing the production of reactive oxygen species [56]. However, this paradigm is shifting, and recent studies show that bacteria can directly interact with host genes that regulate the cell cycle, apoptosis, and other mechanisms of tumor suppression [57].

Helicobacter pylori-associated gastric cancer is a classic example of inflammation-induced carcinogenesis. Prolonged or dysregulated inflammation can lead to uncontrolled cellular proliferation and subsequent tumorigenesis [56]. Several groups have found that H. pylori-associated gastritis and peptic ulcer disease, a common precursor to stomach cancer, are often accompanied by upregulated Th17 activity [58, 59]. Further, H. pylori-infected tissues and associated gastric tumor microenvironments exhibit similarly upregulated Th17 activity and an increase in interleukin (IL)-17 production [60] (Figure 2A). Other bacterial taxa have been associated with increased tumorigenesis accompanied by upregulation of Th17 cells. Enterotoxigenic Bacteroides fragilis (ETBF) enhances colonic tumor formation in a mouse model in an enterotoxin and Stat3/Th17-dependent manner [61] (Figure 2A). This increase of pro-inflammatory Th17 cells in both H. pylori-associated gastric cancer and ETBF-associated colon cancer suggests that certain bacterial taxa have oncogenic potential.

Open in a separate window

Figure 2.

Interconnected relationships of the microbiota and immune system to affect cancer (A), immune checkpoint inhibitor efficacy (B), and colitis (C).

A line with an arrowhead indicates a positive association. A line with a perpendicular end indicates a negative association. A straight line indicates context- or species-specific association. An organism with no line indicates mechanism not yet elucidated.

A) Enterotoxigenic Bacteroides fragilis (ETBF), Helicobacter pylori, and Fusobacterium nucleatum have been shown to increase expression of inflammatory cytokines and/or host oncogenes resulting in cancer promotion [58, 62, 139]. Bacteria from the family Enterobacteriaceae have been detected in the pancreatic adenocarcinoma microenvironment [18].

B) Alistipes shahii, Akkermansia muciniphila, Enterococcus hirae, Faecalibacterium prausnitzii, Bifidobacterium spp., are associated with beneficial immunophenotypes in the context of immune checkpoint inhibitor therapy. Bacteroides spp. have mixed effects; they have been cited as detrimental, different strains enriched in both responders or nonresponders depending on the species, or beneficial. Ruminococcus spp. are associated with beneficial immunophenotypes in the context of therapy. Blautia obeum is associated with a detrimental immunophenotype [20, 83, 84, 85, 88, 89, 90].

C) Bacteroides fragilis, Burkholderia cepacia, and Bifidobacterium spp. may be protective against colitis [20, 104, 133]. Blautia obeum is associated with an increased risk of developing colitis. Different OTUs identified as Faecalibacterium prausnitzii and Bacteroides spp. were enriched in patients who did and did not develop ICI-associated colitis [85, 94]. Created with BioRender.

Inflammation is not the only mechanism by which bacteria can contribute to tumorigenesis. Fusobacterium nucleatum has been shown to induce inflammation and upregulate host oncogenes through interactions between its virulence factor FadA and E-cadherin expressed on epithelial cells, including colorectal cancer cells [62]. This binding triggers clathrin-mediated endocytosis of the bacterium, leading to phosphorylation of β-catenin and subsequent activation of NF-κB, which upregulates inflammatory cytokine production. Downstream activation of WNT, Myc, and CyclinD1 promote cellular proliferation and override cell cycle checkpoints (Figure 2A). Inhibition of clathrin prevented internalization and NF-κB activation, but the oncogenes were still upregulated [62]. This illustrates that bacteria can promote carcinogenesis simultaneously in multiple ways; inhibition of a single mechanism may be insufficient to treat or prevent disease.

Enrichment of certain bacteria has been identified in tumors without elucidated oncogenic mechanisms. For example, bacteria from the Enterobacteriaceae family were found by 16S rRNA gene sequencing and fluorescence in situ hybridization in 76% (86/113) of pancreatic adenocarcinoma tumor samples relative to 15% (3/20) of normal pancreas control samples [18]. In a mouse model of subcutaneous colon carcinoma, bacterial presence within tumors led to increased tumor size, and antibiotic treatment enhanced the efficacy of the chemotherapy drug, gemcitabine [18] (Figure 2A). It is unclear whether bacteria are present prior to cancer development or whether they colonize opportunistically, but local bacterial presence is an important consideration in cancer diagnosis and treatment. These data suggest the existence of a local tumor microbiota in certain cancers that can modulate therapy and also that the bacteria-immune interaction may be adjusted to fight cancer.

In addition to the presence of specific bacteria leading to immunological consequences connected to cancer, large scale case-control studies have demonstrated an association with frequent antibiotic courses in the ten-year interval prior to certain cancer diagnoses [63, 64]. Because cancer can affect the immune system for an interval prior to diagnosis, these authors excluded all antibiotics administered in the year prior to diagnosis, again suggesting that long-term bacterial dysbiosis could affect cancer predisposition [63]. More mechanistic studies are required to determine whether this association is driven by gastrointestinal dysbiosis or local, tumor-promoting communities and/or whether it is a direct oncogenic effect of antibiotics in humans. Animal studies have demonstrated that certain antibiotics enrich for pathobionts capable of stimulating the immune system, leading to bacterial translocation into the mesenteric lymph nodes and disruption of the local inflammatory balance [65, 66, 67]. Certain antibiotics have also been shown to directly inhibit immune cells [67], but the mechanistic interplay among specific antibiotics, the immune system, and the microbiota is not fully elucidated.

Cancer immune therapy

Immune checkpoint inhibitors (ICIs) are monoclonal antibodies that block inhibition of T cell function. ICIs have been developed to block programmed cell death protein 1 (PD-1; nivolumab, pembrolizumab), programmed death ligand 1 (PD-L1; atezolimumab, avelumab, durvalumab), or cytotoxic T lymphocyte antigen 4 (CTLA-4; ipilimumab). The inhibition of these proteins, which negatively regulate immune function, enhances the cytotoxic activity of T cells to increase anti-tumor immunity [61]. The proteins natively suppress T cells at different steps; therefore, inhibitors each have a unique mechanism of action. CTLA-4 is an inhibitory receptor found on T cells that competes with cluster of differentiation (CD)28, a constitutively expressed receptor critical for T cell activation [61]. CTLA-4 is not constitutively expressed on resting T cells but is upregulated following T cell activation [68, 69]. Therefore, CTLA-4 plays an important role as an “off” switch to suppress immune activity, but it can detrimentally affect control of cancer growth by inhibiting cytotoxic CD8⁺ T cells. Antibody blockade of CTLA-4 prevents this inhibition, enabling anti-tumor cytotoxic lymphocyte activity to resume. Similarly, PD-1 also plays an important role in immune suppression. It is expressed on activated T cells, and upon binding to PD-L1 or PD-L2 on the surface of various immune cells, it inhibits T cell activity [70]. Like CTLA-4, it is upregulated upon T cell activation, particularly in regulatory T cells (Tregs) [61]. The PD-1 signaling pathway is essential for preventing autoimmune disease by limiting peripheral T cell activity [70]. Although these two pathways vary somewhat in the mechanisms by which they inhibit T cell activation, both separately and combined, they have been demonstrated to be valuable targets for ICIs in the control of cancer progression [71, 72, 73, 74].

Fundamentally, all ICIs upregulate immune activity that may have been suppressed by Tregs and other immunomodulatory cells. Tregs inhibit immune activity, which can be particularly detrimental in the tumor microenvironment where they have been implicated in the suppression of the anti-tumor inflammatory response that leads to immune tolerance of cancer [61, 75]. However, Treg activity can be beneficial in mouse models of microbially-induced cancer, suggesting that context and the local microbiome-immune milieu is crucial [76, 77, 78, 79]. The gastrointestinal microbiota is essential for the induction of Tregs that tolerate commensal organisms. In germ-free mouse models, induction of Tregs is significantly impaired, but it can be restored by colonization with a defined, nonpathogenic microbial consortium [80]. Furthermore, short-chain fatty acids (SCFAs), abundant metabolites produced by intestinal microbiota, have been demonstrated to be important drivers in the induction and activity of Tregs [81]. Overall, this evidence establishes the intimate relationship shared between commensal microbiota and the immunomodulatory activity of Tregs. This implies that the microbiota has the potential to act as a complicating factor in the ongoing battle among cancer, the immune system, and ICIs by acting on the same immune cells synergistically or antagonistically.

Microbiota and immune checkpoint inhibitor response

Recently, several groups have begun to characterize the role of the microbiota in cancer immunotherapy treatment outcomes to better understand the dichotomy between responders and non-responders (Figure 2B). Anti-IL-10R/CpG therapy, which works synergistically with PD-1 inhibitors [82] to allow proliferation of tumor-specific CD8⁺ T cells, is rendered less effective in the presence of antibiotics, thereby providing evidence that the microbiota plays a crucial role in modulating immune therapy efficacy in a mouse model [83]. Conventional mice treated with vancomycin, imipenem, and neomycin in drinking water for three weeks before tumor injection showed decreased tumor necrosis factor alpha (TNF-α) secretion and increased tumor volume [83]. Further suggesting a role of the microbiota, administration of probiotic Alistipes shahii via oral gavage restored TNF-α production in conventional mice, a promising proof-of-concept supporting the therapeutic potential of probiotics (Figure 2B) [83]. TNF-α was similarly low in tumor-bearing germ-free mice, again implicating the commensal microbiota in the anti-cancer immune response. Germ-free mice were not treated with antibiotics, so the additional effect of antibiotics in the absence of the microbiota is not clear. It is also unknown whether the replenishment of TNF-α levels was sufficient to restore the efficacy of the immune therapy to reduce tumor volume and increase survival. TNF-α secretion was not restored until four weeks after antibiotic therapy, suggesting there is an important pre-chemotherapy window in which to consider microbial composition during which antibiotics might be particularly detrimental.

ICI efficacy can be manipulated by altering microbiota composition [20, 84]. Anti-PD-L1 efficacy differed in C57BL/6 mice from two different vendors; although these mice are considered to be genetically identical, they have unique microbiota compositions, which the authors hypothesized explained the differential efficacy that was observed [50, 84]. Further implicating the microbiota, co-housing mice or orally gavaging feces between groups to obtain similar microbiota compositions reduced the differential treatment response between mice [84]. Alternatively, a transferred, soluble metabolite could manipulate gastrointestinal microbiota composition indirectly through an altered immune milieu. Further evidence supporting the microbiota as a factor mediating tumor regression was the correlation with Bifidobacterium spp. abundance determined by 16s rRNA sequencing and increased tumor infiltration by CD8⁺ T cells, a hallmark of anticancer activity. Administration of a Bifidobacterium spp. cocktail was sufficient to significantly increase the anti-PD-L1 efficacy in the non-responder group of mice [84] (Figure 2B). In a similar study, mice were colonized via FMT from patient samples, followed by tumor inoculation, and then treated with anti-CTLA-4 therapy [20]. They found that donor stool could be clustered according to taxonomy using principal component analysis (PCA), and the cluster correlating with highest relative abundance, Bacteroides spp., was most predictive of tumor decrease after anti-CTLA-4 treatment. Although Bacteroides spp. abundance correlated with response, a causal relationship could have been solidified following probiotic administration of a bacterial consortium. The correlation between microbial abundance and ICI response provides exciting avenues of research interrogation and suggests a possible intervention to adjust the patient microbiome to influence cancer therapy outcomes.

These preclinical studies invoked closer examination of human cohorts and lead to a series of prospective studies in search of microbial and immune signatures that could accurately predict patient response to treatment. Patients with the longest overall survival (>18 months) after ipilimumab treatment for metastatic melanoma had gut microbiota strongly enriched for Faecalibacterium prausnitzii, Gemmiger formicilis, and butyrate-producing bacterium SS2–1 [85] (Figure 2B). They found that certain baseline immune signatures, particularly lower frequencies of Tregs and α4⁺β7⁺-expressing T cells (an integrin which directs T cells to the intestine during periods of inflammation), are correlated with long term benefit (Figure 2B) [85, 86]. It is plausible that the baseline decrease in Tregs systemically could imply a local tumor-microenvironment with fewer Tregs that could be more easily be inhibited by ipilimumab [87]. Finally, in a PCA identifying different microbial compositions between responders and non-responders, the authors found significantly increased inducible T cell activation in response to the CTLA-4 blockade [85]. This study was the first in-depth analysis of the relationship among the microbiota, immunotherapy efficacy and toxicity (discussed in the following section), but the cohort size was small, only one type of cancer and drug therapy combination was examined, and the study was strictly observational without any recapitulation in an animal model. It is thus promising to link microbiota composition to ICI response (Figure 2B), and several clinical trials are currently investigating this relationship (Table 1).

Less than a year later, three large scale studies characterized the relationship between the gut microbiome and immune efficacy [88, 89, 90]. These three landmark papers have been the subject of several reviews by the authors as well as by third parties [4, 6, 7, 91, 92]. Importantly, all three studies identified key bacterial species associated with responders and non-responders (Figure 2B) and immunophenotypes associated with responders, and recapitulated these findings in a mouse model.

Routy et al. evaluated the microbiome and outcomes of a cohort of 100 patients diagnosed with either non-small cell lung cancer or renal cell carcinoma being treated with PD-1 blockade [88]. They discovered that responders had increased microbiota diversity that was enriched for Akkermansia muciniphila and Enterococcus hirae, among other species within the Firmicutes phylum (Figure 2B). Upon FMT of responder and non-responder stool samples into mice, they recapitulated clinical outcomes following tumor inoculation and anti-PD-1 treatment. To illustrate the role of the microbiota, mice that received a non-responder microbiota were supplemented with A. muciniphila and E. hirae, resulting in a favorable anticancer immune phenotype characterized by an increased CD4⁺ T cell: Treg ratio (Figure 2B) [88]. This study illustrates the correlation among microbiota, cancer immune therapy efficacy, and local immune response, and demonstrates the possibility of microbial supplementation to augment ICI activity. Gopalakrishnan et al. conducted a similar study in a cohort of 43 patients being treated with PD-1 blockade for metastatic melanoma. They too found responders had increased microbiota diversity compared to non-responders [89]. However, rather than identify species-specific enrichments between the strata, they characterized responders to have higher proportions of taxa belonging to the order Clostridiales (phylum Firmicutes) compared to non-responders, which had gut microbiota dominated by Bacteroidales (phylum Bacteroides) (Figure 2B). Mice colonized with responder fecal microbiota had significantly lower tumor growth accompanied by higher CD8+ tumor infiltration, an immune phenotype consistent with responder patients (Figure 2B) [89]. Matson et al. identified distinct microbiota compositions driven by increased Bifidobacterium longum proportions in responders within their cohort of 42 patients receiving anti-PD-1 or anti-CTLA-4 treatments for metastatic melanoma (Figure 2B) [90]. In agreement with Routy et al., A. muciniphila was detected in four patients who were all classified as responders, but it was undetected in the other survivors and therefore it was not considered a significant influencer of response in this cohort [88]. Matson et al. also found unique microbial signatures associated with modulating the speed of tumor growth upon introduction into a mouse model in the absence of ICI treatment but did not describe which key taxa were associated with fast or slow tumor growth. Mice transplanted with responder microbiota had significantly smaller tumors than mice transplanted with non-responder microbiota even in the absence of PD-1 blockade. This finding implies an anti-cancer effect of the specific bacteria-immune interaction. Mice colonized with responder microbiota had increased relative abundance of Collinsella aerofaciens, B. longum, Lactobacillus spp., Klebsiella pneumoniae, and Parabacterioides merdae, and their tumors were characterized by increased CD8⁺ T cells and IFNγ production (Figure 2B) [90].

Each of these studies identified different microbial signatures, with some overlap between them, that were correlated with ICI efficacy (Figure 2B). This may be explained by differences in cohorts (e.g. cancer diagnosis, co-morbidities, prior or concurrent exposure to antibiotics) or experimental design (e.g. microbiota analysis pipeline, production of humanized mice, in vivo cancer model). Alternatively, these findings could imply that different bacterial species or genera are accomplishing the same metabolic role or that some unifying feature (e.g. membrane composition, lipopolysaccharide, or other molecular signal) is being recognized by the gut mucosal immune system. These data encourage the use of deeper sequencing methods, such as shotgun metagenomic sequencing, as was done by some of the studies described above, instead of 16S rRNA sequencing, and they highlight the utility of performing metatranscriptomics, metaproteomics, and metabolic pathway analysis (Figure 1) to identify unifying signatures indicative of protective and detrimental microbiota.

Microbiome and immune checkpoint inhibitor toxicity

By removing inhibitory checkpoints on immune cell activity, ICIs activate effector T cells to target tumors, but have adverse effects against numerous other tissue types [93]. The most common sites of adverse effects (gut and skin) are characterized by rapid cellular turnover and have close association with bacteria, suggesting a role of the microbiota in toxicity. The relationship between the microbiota and host biology is particularly complex in the context of inflammation. Microbes can modulate inflammation, which can in turn shape the microbiota, leading to a positive feedback loop characterized by dysbiosis and disease. Although there is strong evidence that both the microbiota and the immune system play critical roles, it is unclear which is the driving force. It is likely a combination of both, with dynamics that vary with host background, microbial composition and function, and cancer diagnosis and treatments.

Up to 25% of patients treated with ipilimumab (CTLA-4 inhibitor) experience colitis, so clinicians and researchers have undertaken studies to identify associations between the microbiota and toxicity [94]. In their 2015 mouse experiments, Vétizou et al. showed that toll-like receptor (TLR) agonists (i.e. the microbiota), intraepithelial lymphocytes, and ICI treatment were all required to induce apoptosis and ICI-associated colitis [20]. Further, they demonstrated that a prophylactic gavage of Burkholderia cepacia and Bacteroides fragilis provided protection against colitis during anti-CTLA-4 treatment (Figure 2C). To the best of our knowledge, this was the first demonstration, not only that ICI-associated colitis was microbiota-dependent, but that prophylactic probiotic treatment can protect against colitis in mice [20].

Human cohorts have also been studied to determine whether gut microbiota composition could predict or modulate colitis in addition to ICI efficacy. In a prospective study, Dubin et al. obtained fecal samples at the start of ipilimumab therapy for metastatic melanoma and stratified patients who did or did not develop colitis thereafter [94]. The authors used both 16S rRNA and shotgun sequencing to identify an increased relative abundance of Bacteroidetes in patients who did not develop colitis (Figure 2C). Similarly, they identified functional pathways using HUMAnN to show polyamine transport and riboflavin, pantothenate, thiamine, and biotin biosynthetic pathways were enriched in those protected from colitis [39, 94]. It is unclear whether these pathways are simply Bacteroides-associated functional pathways, directly inhibit cytotoxic activity of T cells, or otherwise provide local immunomodulatory effects, but it is intriguing to be able to utilize patient stool at therapy onset to predict toxicity.

In a similar cohort, increased abundance of Bacteroidetes relative to Firmicutes correlated with a longer colitis-free therapy period [85] (Figure 2C). However, the observed decreased incidence of colitis was offset by decreased survival. The authors initially show that ipilimumab, a CTLA-4 antibody, did not affect microbiota composition at the phylum level during treatment in the absence of colitis. They also conclude that ipilimumab did not decrease Shannon or Simpson diversity throughout treatment, but because of the small numbers of patients (only four patients sampled at the end of therapy), the null hypothesis of no difference between timepoints could be secondary to high variance in microbial composition. During colitis, however, they found decreased Shannon diversity as well as specific reductions in members of the Firmicutes. These authors proceeded to show that in addition to better clinical response, increased relative abundance of Firmicutes at baseline correlated with colitis. In a linear discriminant analysis to identify taxa associated with colitis, two OTUs designated as Faecalibacterium prausnitzii were strongly correlated with differential response (Figure 2C). This discrepancy, which may be artifactual based on OTU picking or may represent different metabolic contributions, highlights the importance of identifying strain level differences using ASV-based methods for 16S rRNA sequencing [31, 34] or with shotgun metagenomic sequencing [41, 95]. To explain the balance between ICI response and toxicity, they performed flow cytometry on whole blood at baseline and correlated immune populations with microbiota composition. They found that patients who ultimately developed colitis had decreased baseline levels of blood IL-6, IL-8, and sCD-25 relative to those that did not develop colitis [85] (Figure 2C). IL-6 and IL-8 are pro-inflammatory cytokines and sCD-25 is important for IL-2 function in promoting Treg development [96]. Paradoxically, patients with a pro-inflammatory state at baseline had lower incidence of developing colitis during treatment. Although preliminary and requiring further replication, it is interesting that microbial and immune biomarkers potentially exist that could stratify patients at the start of treatment such that probiotics or fecal microbiota transplant could alleviate or lessen toxicity while maintaining appropriate therapeutic efficacy.

Modulation of the microbiota to improve therapeutic outcomes

Future directions

Baseline immunologic, gut microbial taxonomic, and functional pathway abundance could be used to develop personalized cancer immunotherapy treatments to maximize progression-free survival while minimizing toxicity. In the near future, fecal samples could be sequenced by clinical and molecular microbiology labs and directed microbiome adjustments could proceed simultaneously with or prior to initiating ICI treatment to augment response and limit toxicity. It is therefore vital to maximally replicate the findings described above and elsewhere in the literature in different cohorts with numerous analytical techniques (Figure 1). Studies are currently ongoing to identify microbial and immune differences correlating with toxicity and response to ICI (Table 1). Research studies with apparently conflicting taxonomic correlations discussed above could be related to insufficient resolution or different metabolic contributions of strain-level differences among the same species [134, 135]. We encourage researchers to consider incorporating functional metabolic profiling and strain tracking into longitudinal analysis to more precisely define the changes accompanying ICI therapy, toxicity, and response to microbial intervention.

In addition to potential microbiome adjustments with prebiotics, probiotics, postbiotics, antibiotics, and FMT, ample novel therapeutic strategies are emerging. An exciting technique would be to selectively kill pro-oncogenic or pathogenic bacteria with specific bacteriophages as has been recently achieved in other immunocompromised populations [136]. Genetically engineered microbes can also be used in vivo to diagnose and treat cancers [137]. It is compelling to consider other non-bacterial members of the microbiota including viruses and fungi and their role in modulating microbial composition and immune function as well as producing anti-cancer metabolites [138]. The era of personalized medicine is fast approaching due to the decreased cost of next-generation sequencing and the increased understanding researchers and clinicians share regarding the importance of the gut microbiome in regulating the immune system and disease.

Conclusions

Understanding microbiome compositions, manipulations, and interventions is now vital to understanding cancer progression and immunotherapy-based treatments. The details underlying the intricate relationship among the host immune system, cancer progression, the commensal and tumor microbiota, and therapeutic interventions continue to emerge. In addition, the tools and techniques used to characterize the microbiome are continually evolving to provide more data at lower costs. Although we have collectively learned much about commensal microorganisms and the modulatory role in local and systemic homeostasis, novel cancer immunotherapies have brought microbiome correlations and manipulations to the forefront. To progress beyond mere association in order to elucidate causation, we need to characterize precise molecular mechanisms by which the microbiota mediates cancer immunotherapy response and toxicity. Only after these molecular mechanisms are uncovered can interventions to modulate the microbiome to affect response and toxicity be widely employed. Continued partnerships among clinical scientists, laboratory physicians, and other translational researchers will help to usher in the era of personalized medicine with improved outcomes of many diseases while limiting often intolerable side effects.

Acknowledgements

Abbreviations

Footnotes

References